Open this documentation inside the application · Français
Room Correction Helper (RCH) is a calibration tool for Denon and Marantz home theater systems based on Audyssey. It works with REW to measure, align speakers and subwoofers, generate filters and verify previews, then transfers the calibration directly into your AVR through the RCH Bridge, a small local program that drives the receiver over the network (Audyssey measurement included).
RCH is not a one-click magic button. The goal of the application is to provide a guided but transparent process so you can verify each step before the final transfer.
This documentation describes the current behavior of the application and follows the commands visible in the interface.
This help page is integrated directly into the application. The instructions below assume that you are already inside RCH and can switch back to the Application tab at any time.
The reverse path exists too: every tool zone of the application carries a
Documentation link that opens the section describing it. Those links are
written into the page address as #documentation/autoeq: an address
copied from here therefore reopens the right section, and can be shared as is.
Under each group of buttons, one sentence recalls what the tool reads, what it
computes and where it writes; the full detail stays in the button tooltip.
Before using RCH, make sure the following points are covered:
RCH needs three links to work, shown at the top of the Application page as three indicators:
As long as the chain is incomplete, the application is deliberately locked and the banner lists the missing links. This is by design — the receiver configuration (model, MultEQ version, detected channels) is read live from the connected AVR; failing that, it has to come from an imported session.
Each step is locked on what it actually needs, not on the whole chain:
Browser note: recent Chrome and Edge ask for a "local network" permission the first time RCH connects to the bridge — accept it. Firefox works without any extra step. Safari is not supported with the https site (WebKit limitation): use Chrome, Edge or Firefox.
Practical advice: keep a clean working folder with your measurement files (MDAT, ADY, MQX, liveproject), your RCH session files and your target curves. This avoids many file-handling mistakes.
If you use a custom target curve, prepare it in advance as well. You will need it in REW for Align SPL and for final validation.
Whichever starting point you pick, the receiver configuration (model, MultEQ version, amp assignment, detected channels) comes from the AVR, read through the bridge — never from a measurement file. The .ady, .mqx, .liveproject and .mdat files only provide measurements. A session file is the exception: it carries a copy, taken on the receiver owner's machine.
The AVR configuration panel shows that configuration. It appears as soon as the configuration is loaded, bridge connected or not: an imported session is enough. When the receiver is not reachable, the panel says these values are the ones captured in the session, with the date they were read. That date is what qualifies them.
Part of what it shows does not come from the receiver. The model, the MultEQ generation, the subwoofer mode and layout, the detected channels are its answers. The speed of sound, the MultEQ filter lengths and the model capabilities are not: RCH derives them from those at capture time, and the session keeps them as they were. That derivation has already been corrected. An old session can therefore carry values RCH would compute differently today. They are shown as captured, never recomputed, so that the card does not diverge from what the file holds.
The two most common starting points sit side by side on the Application page: let the bridge measure for you, or drop a measurement file you already have.
The bridge drives the receiver to play the Audyssey sweeps on every speaker, and RCH imports each impulse response into REW as the measurement progresses. This is the most direct path: no file to handle, and the measurements arrive already named by channel and position. The full walkthrough is in Measuring with the assistant.
If you already measured with the Audyssey MultEQ Editor app (.ady), with MultEQ-X (.mqx) or with Dirac Live (.liveproject), drag and drop the file onto the drop zone of the Application page. The measurements are loaded automatically into REW.
These files provide measurements only: the configuration already loaded does not change. If the file does not match that configuration (different model, different MultEQ version, different subwoofer mode, different subwoofer count), RCH shows a warning but lets the import proceed — it is up to you to judge whether the measurements are still relevant for the receiver in place. None of these differences refuses the import: the configuration does not come from the file, which only testifies to the conditions the measurements were taken in.
If you measure manually in REW with your own mic, your measurement names must follow the Audyssey names of the detected channels of the receiver (visible in the AVR configuration panel), otherwise they will not be matched to the correct channels in the measurement table.
The most reliable sequence is: start REW empty, enable its API, connect the full chain, and only then open the MDAT file in REW. Before calculating, also make sure your measurements are not carrying leftover EQ, delay, or inversion state.
Before the measurement session, return the AVR to a simple and predictable state:
On the REW side:
Measure every channel with the played sweeps, preferably several times and at several positions. Measure the main listening position last and leave the microphone in place: post-calibration checks will be easier.
RCH can also import Dirac Live projects exported as
.liveproject. Load the file just like an ADY (drag and drop, or click
the drop zone on the Application page). As with ADY, the automatic import into REW
only happens when RCH is connected to REW.
Unlike an ADY, a .liveproject does not contain ready-to-use impulse
responses. RCH reconstructs the true impulse responses (with phase
and inter-channel delays) from the raw microphone recordings stored in the file, by
deconvolution. The microphone calibration embedded in the project is
already applied during reconstruction: do not load a microphone
calibration file into REW for these measurements, or you will double-correct the
treble.
The file may contain several measurement positions; RCH recovers all of them and names each measurement by channel and position. This processing is heavier than an ADY import, especially for a multi-position project: it runs in the background and can take from a few seconds to about a minute. Progress is shown in the log.
The relative levels and delays between channels are preserved (the closest speaker sits at time zero, the others offset by their real delay). The absolute distance is not recoverable, and subwoofers appear farther away because their delay includes the subwoofer DSP latency: this is normal.
Important: like the other files, a .liveproject only provides
measurements. The configuration (channels, MultEQ version, options) stays the
one RCH already has: the AVR connected through the bridge, or an imported
session. That is what drives the optimization and the final transfer.
The Audyssey Measurement section of the Application page drives a complete measurement session through the bridge: the receiver plays the Audyssey sweeps on every speaker, and RCH imports each impulse response into REW as the session progresses, already named by channel and position. The Audyssey microphone calibration is handled by the bridge: you have no calibration file to load.
The assistant is optional: you can just as well measure yourself in REW with your own microphone (see Choose your starting point, advanced mode) or import an existing measurement file.
Then click Start measurement session. If a precondition is not met (microphone not plugged, headphones plugged, receiver powered off...), RCH shows an explicit message: fix it and retry.
For each detected subwoofer, the Subwoofer level matching block shows a live SPL reading: click Start level-matching signal, turn the gain knob of the subwoofer until it reads about 75 dB, then click Stop. Levels are matched one subwoofer at a time: while a subwoofer is being adjusted, only that one plays — this is normal, not a fault or a wiring problem. The block only appears if the receiver supports subwoofer level matching; otherwise, set each subwoofer gain from the receiver's own menu.
This step comes before position 1, and only before: once the detection pass is done the receiver refuses the routine, and RCH closes the block and says so. To redo it, complete or cancel the session and start a new one.
Position 1 always measures the full channel plan detected by the receiver: it is the detection pass (speaker presence, polarity). Channel selection only applies from position 2 onward: you can then remeasure only a subset of channels if needed.
For each position: move the microphone, click Measure position, and wait for the sweeps to finish (progress is displayed). The responses are imported into REW as the measurement progresses. Any warnings (reversed polarity detected, asymmetry) are displayed but do not block the session: note them and check the physical wiring if an inversion is reported.
Re-measuring. The current position — the last one measured — stays available: the Re-measure position button sweeps it again. Untick the channels you do not want to redo to re-measure a single speaker. The measurements concerned are replaced in REW, those of the channels not swept again stay in place, and RCH names the position before replacing anything. Position 1 is the exception: it is the detection pass and always re-measures the full detected plan.
Only the current position can be re-measured, and that is deliberate. In REW, the position of a measurement is its rank in the list for its channel: a re-imported measurement lands last, hence exactly in its own place when it is the last position — and it would shift every later one if it were an earlier position. An average has no rank: it does not count in the position of the takes, it shows as “Average”, and its sums and previews carry the “avg” mark (LFE predicted_Pavg). To redo an older position, measure it as an extra position, then delete the old one in REW if you no longer want it.
When all positions are measured, click Complete session to hand control back to the receiver. Cancel session abandons the session at any time.
Then save your measurements in REW (File / Save all measurements) as a .mdat file: that file is what keeps your measurements from one working session to the next — RCH stores no signal data (see Session).
The workflow has ten steps, to be followed in that order: each one builds on the result of the previous one, and going back makes what follows stale. Find Sub Alignment aligns the subwoofers over the full band, without a crossover: changing a crossover at step 9 therefore does not send you back to step 8. Only an alignment from an earlier session, computed at a crossover, may still ask for one return to step 8 when that crossover changes. The new Find Sub Alignment run then replaces it with a full-band alignment, which a crossover change no longer makes stale.
Steps 1 to 5 deal with the speakers, steps 6 and 7 with the subwoofers among themselves, steps 8 and 9 with the handoff between the two, step 10 with the transfer. Steps 6, 7 and 10 have their detailed instructions elsewhere — Subwoofers and MSO and Finalization and transfer: what is said about them here gives their place in the order, not the detail of every button.
Every step opens with four lines: what it produces, the gestures to perform, what to look at, and the state it leaves the project in. What follows those four lines gives the reasons and the pitfalls.
The Guided journey ribbon, at the top of the Application page, carries these ten steps, and each chip states where its own step stands. A chip that changes colour and takes the ⚠ sign shows the stale state: one of the step's inputs changed since it was last run here. Setting a crossover by hand, recomputing a filter or changing the target curve is enough. It is not an alert, and RCH recomputes nothing on its own: the step card names the cause under its title, and you are the one who decides whether to run it again or carry on. The one mandatory return is said elsewhere — the strip above the table, which sends you from step 9 back to step 8.
Below the ribbon, the step card carries five buttons: Previous step, AUTO, UNDO, Go to the controls and Next step. AUTO does what the step's button does, with the settings shown in its controls: there is nothing more to set. UNDO puts back the state from before the last AUTO, as long as nothing else has been started since; reloading the page forgets it. A line under the four blocks gives the result, or why AUTO or UNDO is not available and what to do. AUTO is offered at steps 2 to 10; step 1 is done by hand. Without a subwoofer, steps 6 to 9 have nothing to do and step 10 has no low-pass to set: the card says so instead of offering AUTO.
From step 3 to step 10, AUTO works on the averages: it shows the Average position before starting the step, and UNDO puts the previous position back. When the step is only partly done, the line says so in orange instead of announcing a success, and gives the cause: an alignment capped by the distance, crossovers set on only some of the groups, a polarity still to be decided. A speaker that the equalizer leaves without a filter, because it already follows the target or because the correction would move it away, no longer stops anything: it stays without a filter, and both the crossover search and the low-pass search take it as it is, as long as the last filter generation left it that way. After step 9, the line gives figures for the predicted system at the working position: deviation from the target, bass level, summation and group delay at the junction.
On the subwoofer side, AUTO does a little more than the button at two steps. At step 4, before Align SPL, it compares the physical level of each subwoofer with the others, as measured, without the trims RCH has already set, on average over the measured positions. Trims only correct a small difference: a large one uses up the gain reserve of the AVR. When a difference goes beyond what measurement noise explains, AUTO asks whether your subwoofer gains are set. Confirm if they are, and AUTO aligns the levels; cancel, and AUTO ends without changing anything and opens the subwoofer level check. At step 7, before Equalize Sub, AUTO sets the level of the subwoofer sum. It reads the dips of the sum within its own passband and keeps only those that rise again on both sides: the roll-off towards the edge of the band is not one. It leaves aside the narrow, deep dips that equalization does not fill, with their flanks, and those the AutoEQ leaves aside because they are deeper below the target than its Notch exclusion setting. Then it moves every subwoofer by the same step so that the deepest of the others is within reach of the equalization; with no dip kept, the middle of the sum is set on the target. A subwoofer trim does not go above +9 dB: above it, the subwoofer output of the AVR would run short of headroom at high volume. The Application logs panel gives that level and what set it. If a trim reaches its limit, if half a decibel more or less would equalize better, or if the sum cannot be read, the line says so in orange. The Equalize Sub button leaves that level to your setting. When the equalization of the sum has nothing to improve, for instance because the joint alignment already set it on the target, it succeeds without setting a filter, and the logs say so.
At the junction, AUTO also decides two inputs the buttons leave to you. At step 8, when no reference speaker is chosen, it aligns the subwoofers on the center, or on the front left without a center, sets that choice in the list and says so in the logs; the button waits for your choice. At step 9, before Find Best Crossover, it computes the localisation ceiling: it tries the crossovers of the front speakers without penalty and without writing anything, and sets the ceiling at the lowest value that lets each of them reach its best crossover within measurement noise. The other speakers then look for theirs with the penalty above it. The setting shows the value and stays adjustable, and the logs say which speaker set it. After the polarity check, AUTO inverts a speaker group whose members all sum better with the subwoofers once inverted, runs Find Best Crossover again for it, and reverts the inversion if the new check contradicts itself; the group the subwoofers are aligned on is never inverted. A channel that clearly prefers the other polarity stops AUTO before the previews: start by checking its wiring. A disagreement within measurement noise does not stop it, and the polarity report shows it. UNDO restores the ceiling and the inversions. The polarity check button itself inverts nothing.
An imported session fills the table without the bridge or the receiver being there. The chain strip tells the truth: in that state, study is possible, transfer is not.
The Speaker layout card, below the AVR configuration panel, draws the detected channels as seen from the listening position, each with its measured distance: it is the quickest way to spot a missing channel or an implausible distance.
In advanced mode, avoid loading the MDAT before connecting. RCH populates its table more reliably when REW is already running, the API is active, and the file is opened once the chain is complete.
Picking up a project you have already worked on is not a step of the workflow: it is the gesture that precedes it. Two commands do that, and they do not do the same thing.
Use Reset Measurements if you want to reset all measurements to their default parameters before running calculations again. Use Reset Application if you want to clear the whole project after a bad workflow, a wrong file, or a complete project change.
In practice, Reset Measurements clears smoothing, IR windows, target settings, room curve, generic equalizer, delays, and inversion state. It is the right first step before repeating a full workflow.
As soon as the table fills in, every row carries a thumbnail of its frequency response. It is there to show at a glance what a measurement is worth, and to let you compare rows: all thumbnails share the same scale, 10 Hz to 20 kHz horizontally and a 35 dB window centred on the target level vertically. A curve drawn higher than another really is louder, and a dip seen on one row really does compare with the dip on the row below. The window being tight, whatever goes more than 17.5 dB above or below the centre runs along the edge of the frame instead of leaving the drawing: the curve stays continuous, and the edge says "it goes past here".
A thumbnail carries up to three curves, all of them read from REW — none is computed by RCH:
The thumbnail follows the smoothing RCH applies: change the smoothing method and every affected thumbnail updates on its own. The same happens after each operation that changes a measurement (Align SPL, Time align, IR windowing, inversion). While it refreshes, the previous curve stays on screen, dimmed, so the column never flickers.
Thumbnails are read in the background and always give way to your workflows: nothing is read while a calculation is running, and the rows on screen come before the others. A change made by hand inside REW is not reflected immediately: the thumbnail catches up on the next operation started from RCH.
Click a thumbnail to open the assessment of the measurement (or reach it with the keyboard and press Enter): its first card is the enlarged view of the thumbnail. RCH then reads the measurement at full resolution — REW's native grid, not the condensed version used by the thumbnail — then its corrected response the same way. The cards that follow are the criteria of the acoustic assessment. In the header of the column, two buttons — Assess the measurements and Compare before / after — grade every measurement of the position on display in one go.
That first card has graduated axes: round frequencies from 10 Hz to 20 kHz, levels every 5 dB. You find exactly the same curves as in the thumbnail, under the same rules, and the same scale — same bounds, same 35 dB window, same centre. A legend names each line, and lists only those actually drawn. The card shows the smoothing applied, the title names the measurement. Close it with Esc, with the cross, or by clicking outside the window.
Two takes made at different times do not average together, even if the curves look alike. The microphone level, the exact position and the state of the receiver are part of the measurement.
Keeping position 1 in addition to the averages is useful if you later want to compare the result at the main listening position without losing the averaged curves.
The three averaging methods are Vector average, RMS + phase avg., and dB + phase avg. If you do not already have a clear preference, keep the same one throughout the project so your comparisons stay meaningful.
Before each average, RCH now time-aligns the positions itself (envelope cross-correlation for the coarse estimate, then a constrained fine cross-correlation), replacing REW's Cross corr align which could slip by a full cycle on some measurements. This alignment improves the coherence between positions, especially for the Vector average and for subwoofers.
Without this alignment, RCH makes no average. It refuses before writing anything, for every channel:
The message says what to do.
For filter calculation, prefer RMS + phase avg.: this method preserves the level at all frequencies. Vector average remains valuable for time alignment and for judging the coherence between positions, but it underestimates the measurement level (by -1 to -2 dB with 3 positions, up to -5 dB with 8 positions): filters computed on it will be under-dosed.
AUTO creates the averages with the strategy and method shown. UNDO undoes them with the Keep Everything strategy, the default, if no average existed before: a deleted original measurement cannot be recreated. The card says so before you click.
The gain set by a previous handoff is reset to zero. That is deliberate: you do not hand over on a time base that is still going to move. If you run Time align again after step 8, steps 8 and 9 have to be done over.
The time reference is no longer the impulse peak but the arrival time of the wavefront, estimated through the excess phase (REW's "Estimate IR delay" method): the peak can land on a reflection stronger than the direct sound and skew the distance by several meters, whereas the arrival stays stable. The resulting distances are therefore closer to the physical ones.
Do not re-run speaker time alignment after this step without a reason: the following steps depend on it. If you do re-run it, the alignment of the subwoofers with each other (step 6) survives: the subwoofer block is re-anchored as a whole, its relative offsets are kept.
Up-firing Atmos speakers (laid flat on the mains, aimed at the ceiling): the useful path is the ceiling bounce, not the small direct sound that leaks toward the microphone. Alignment and distance follow that bounce; check that the distance is consistent on these channels.
The Timing diagnostic columns bar adds the time columns: Delays (ms), Impulse peak time (ms), Absolute Distance (m) and Required Shift (ms), plus the Junction deviation (dB) column. They are hidden by default to keep the table readable: click a chip to show the matching column, click it again to hide it. The choice is kept from one session to the next.
An action that WRITES into one of these columns opens it by itself, so that a result is never written into an invisible column: Find Sub Alignment opens Required Shift (ms) and Junction deviation (dB), while the automatic best-crossover search and the polarity report open Junction deviation (dB), and the per-line recompute button (⟳) does the same for the two columns it rewrites. This only happens as long as you have not spoken: once you have shown or hidden a column yourself, your choice comes first and nothing undoes it.
Junction deviation (dB) is what measures the quality of the speaker-to-subwoofer handoff: how far the resulting response departs from your target curve in the bass, once the speaker and the subwoofers are crossed over. The lower the value, the better the blend. Required Shift (ms) is still shown for information only: it is no longer the criterion RCH uses to choose a crossover frequency or a polarity.
The target curve must be loaded in REW before using Align SPL: it is what the levels are referred to. If RCH cannot find it, load it in REW and run again.
For subwoofers, Align SPL sets the level of each sub — measured with the same method as the Subwoofer level check card — exactly on the target level (shared for the coherent sum). That level is a mean SPL weighted by the junction domain: each frequency counts for the share of the crossovers your amp can offer where the sub still works there. Beyond 250 Hz — the highest possible crossover — the sub will be muted anyway, and whatever it radiates there no longer weighs anything. No passband is detected, so no subwoofer is ever rejected, even one reaching 1 or 2 kHz as many DIY subwoofers do.
Align SPL lowers the target level, and that is not a loss of calibration. RCH asks REW for the level at which the target curve fits the reference speaker best, then applies that level to every measurement: the target lands where your system actually plays. 75 dB before, 70.9 dB after is an ordinary case — the Target measurement carries it in its title. There is nothing to make up on the volume knob. Keeping the target at 75 dB would mean raising every trim by 4 dB: that much headroom spent on the receiver for a result the volume knob gives at no cost, and that is what it is for. That knob is calibrated nowhere, and a microphone measurement carries no absolute level reference — which is why most calibration software puts its reference at 0 dB.
What actually goes to the receiver is the difference between a
measurement's level and the level RCH saw when it arrived — not the level itself.
That difference is the SPL column of the table, rounded to the receiver's
half-decibel step: it is 0 dB at import, whatever the source (.ady
file, bridge, Dirac Live project), because RCH does not know a measurement's
original level and does not guess it.
RCH never writes into REW's SPL offset: every level write it makes goes through Align SPL, and undoes in one gesture in REW with Undo SPL alignment. If you move a measurement yourself with REW's SPL offset, that move is counted in the trim as long as RCH keeps its session. Once it starts from a fresh session — another browser, a cleared cache, measurements imported again — it starts again from the level it reads at that moment, and that manual move is no longer counted. So set levels in RCH rather than in REW.
These settings are not a step of their own: they decide how detailed the filters produced by this step will be, and they are set once for the whole project.
Smoothing and windowing drive the level of detail of the equalization — light smoothing produces detailed corrections with many filters, heavy smoothing subtle corrections with few filters — as well as how the curves are displayed. Subwoofer alignment computations (Align Sub, predicted sum, theoretical ceiling) always measure the raw response, regardless of these settings: your display choices cannot degrade alignment accuracy.
The Optimized MTW windowing, set by default, removes the energy of the reflections in the treble so the filter only corrects the direct sound, and keeps every detail in the bass. The working curve therefore drops in the treble, and RCH lowers the target by the same amount — this is intended, and Windowing: what the filter is allowed to see explains why, what becomes of the dip that appears, and what the choice costs. If you prefer a curve that is detailed in the bass and smooth in the highs without any level change, variable smoothing is the recommended alternative.
In the same spirit, RCH corrects the bass finely and smooths the highs heavily: in the highs the direct sound is already well below the target, and not over-correcting it is the recommended practice. The frequency range you set is authoritative — RCH does not impose any zone. Keep in mind that it is the combined corrected measurement + real chain that must follow the target: avoid over-correction.
The crossover frequency is not an input of this calculation. The equalization corrects the shape of the response towards your target curve without knowing anything about the handoff with the subwoofer: that is why crossovers are chosen further down, at steps 8 and 9, once the subwoofers are frozen. So generate the filters with whatever crossover is already in the table — the receiver's value, or a starting point. What really has to be redone after a crossover change is the subwoofer alignment if the reference speaker is involved (step 8), the previews, and the saved banks (step 10).
Filters are computed by RCH. If you prefer to generate or adjust them directly in REW (EQ / Match target, or manual editing), RCH detects that the filters changed and uses the ones present in REW for the rest of the process: previews, sub alignment and transfer.
The calculation itself is driven by the Settings panel. The defaults are designed to give a good result without touching anything: if you want to understand what happens when you ask for a filter, or fine-tune the behaviour, everything is detailed in Filter calculation (AutoEQ).
Some installations measure a single subwoofer for several declared ones. There is then nothing to align between them. RCH says so under the subwoofer tools, and the transfer card announces that the calibration of the measured subwoofer will be written as is onto the others.
Several subwoofers in a room do not simply add up: depending on the frequency, they reinforce or cancel each other. Where you sit, two subwoofers can give less than one. The alignment looks, for each of them, for the delay, the gain and the polarity that give the best sum at that seat — not in the room in general, which does not exist as a listening point.
At this stage RCH has no crossover information at all, and that is deliberate: the subwoofers align with each other over their own band, independently of the speaker their sum will later hand over to. The handoff is what steps 8 and 9 are for; mixing the two would make the alignment depend on a choice that has not been made yet.
The commands, the joint optimization and the level check card are detailed in Subwoofers and MSO.
RCH does not decide the level of the sum: you set it, you run the equalization again, and you repeat until it suits you.
Going back to that base does not break anything silently: RCH then shows a strip above the table naming what became stale and why, and it recomputes nothing on its own.
Find Sub Alignment compares the reference speaker with the actual sum of the subwoofers, both as they come out of equalization: filters, all-pass filters, levels, each subwoofer's own polarity and the delay differences between subwoofers included. It adds no handoff high-pass or low-pass and reads no crossover. It searches for the common shift, and whether to invert every subwoofer together, that give the largest sum between 20 and 320 Hz: it tries both choices over the same delay range. Filters, all-pass filters and the delay differences between subwoofers are left unchanged. An inversion flips every subwoofer's polarity, but the polarity relationships between subwoofers stay the same: two subwoofers of opposite polarity remain so.
That range comes from the distances of every channel, speakers and subwoofers. Speakers do not move; no subwoofer distance can go negative or exceed what the AVR accepts. When the best sum lands on a limit of the range, RCH says so: it is the best within that range, and a better position may exist beyond it. To widen the range, reduce the distance gap between the closest and the farthest channel, then measure again.
If the responses do not allow the sum to be evaluated, if a distance is missing, or if the distances leave no range at all, RCH stops before writing anything and names the cause. After Time align, the log already announces the distance adjustment Find Sub Alignment will make and whether it will flip the polarity; the former announcement of a distance shortfall, in metres, no longer applies to this calculation.
The final message and the log give the distance adjustment, the matching delay, the full-band summation gain at the reference position, and the range limit when it came into play. The log also says whether the subwoofer polarity was kept or flipped, then the resulting state: normal, inverted, or different per subwoofer when the optimizer set them that way. That gain is the best the calculation found, not a promise: it guarantees neither a gain at the other positions, nor the best sum once the handoff is in place, at every crossover. Step 9 — previews and polarity check — is what judges the real handoff.
Find Sub Alignment is also what decides the subwoofer polarity, once and for all, on the reference speaker you selected: it keeps or inverts the subwoofers together, whichever gives the larger full-band sum. There is no decision threshold; on an exact tie, RCH keeps the polarities in place. The 2.69 dB threshold remains that of the polarity check at the handoff, at step 9, which reports without inverting anything.
The same shift and the same polarity choice apply to the subwoofers of every position. If the operation stops while writing, RCH puts every subwoofer it touched back in its previous state, and the message names any that could not be.
The other speaker groups follow that decision. If one of them would prefer the other polarity, RCH reports it instead of inverting the speaker: inverting a channel is a wideband remedy for a narrowband problem, and it also changes that channel's polarity relative to every other channel.
Find Sub Alignment does not touch LPF for LFE. That setting has its own button and is chosen at finalization (step 10), once the crossovers are settled: it is not a by-product of the alignment.
For every candidate frequency, RCH builds the resulting response — the speaker and the subwoofers once crossed over at that frequency — and compares it to your target curve between 20 and 320 Hz. What it measures is a deviation in shape: the overall level is removed from the calculation, because SPL matching is not the crossover's business. The frequency it keeps — and selects in the list for you — is the one whose response comes closest to the target, averaged over the speakers of the group and over the measured positions.
Why this criterion rather than the timing offset at the junction: the handoff RCH builds is a symmetric LR24|LR24 pair (see the note at the end of this step), so it is exactly neutral. It degrades nothing: it only chooses where the handover happens between two already fixed responses, the equalized speaker and the subwoofer sum. The old criterion measured the timing residual, a few tenths of a dB, and was blind to the difference between those two responses, which is 3 to 5 dB — and that is the real subject.
Several crossovers are often equivalent, and RCH says so instead of pretending to decide. In practice, the gap between the best frequency and the next one is less than a tenth of a dB, far below the measurement noise. RCH estimates that noise from the positions of your own session and declares indistinguishable every candidate that falls within it. Inside that group of equivalents, it keeps the highest one: it is the one that demands the least excursion from the speaker, and bass localisation is already handled by the penalty below.
Above 120 Hz, the limit beyond which bass starts to be localisable, a candidate is not forbidden but it is penalised by 2 dB per octave: 0.64 dB at 150 Hz, 1.47 dB at 200 Hz. A high crossover can still win, but it has to earn it with a real gain in response; in practice that happens in fewer than one group out of ten.
Lower bound: RCH first measures how low the speaker really holds its level — within 3 dB, over a full octave — and allows no crossover below that. There is no point asking a speaker to reach down where it no longer plays.
The log details, for each tested frequency, the deviation from target, any penalty, membership of the indistinguishable group, and the reason for a rejection (below the speaker's capability, missing data).
If, when listening, you consistently prefer lower crossovers, the lever to move is the localisation ceiling — not the selection rule: that ceiling, and it alone, is what sets the preferred crossover. It is adjustable in the Localization ceiling card (filter settings): bounds 60 to 150 Hz, 120 Hz default restored in one click, value kept in the session. Candidates above the ceiling are still evaluated and penalised (2 dB per octave), never forbidden, and the decision sentence states the ceiling that was used.
Because the crossover is a group property, the per-line button only appears on the first speaker of each group. Clicking it processes the whole group.
Requirements, already met if you followed the workflow: the speaker filters must be generated (step 5) and the subwoofer sum must exist. Preview Sub is what produces it, titled LFE predicted_P1 — one projection per position. Without it, the per-line button shows an alert and does nothing; the header button, for its part, logs the failing group and carries on with the other groups.
This search no longer inverts the polarity of your speakers. Polarity is decided once, globally, and it applies to the subwoofer: Find Sub Alignment is what sets it (step 8). If a group would prefer the other polarity, RCH reports it in the log without inverting anything.
The search also judges the entire opposite-polarity branch: if, with the group speakers inverted, another crossover would do clearly better, the log sentence says so — the crossover that branch would retain, its deviation from target, the gain (localisation penalty included) — and the audit table shows the opposite-polarity deviation for every candidate. Nothing is applied: if you decide to follow the signal, invert the group speakers together (every position) and run the search again. Do not re-run Find Sub Alignment for this: it would flip the subwoofers for every group. The sentence also tells the likely cause apart: a preference that changes with frequency is a phase rotation (not a wiring fault); a uniform preference deserves a wiring check first.
Find Best Crossover writes the crossover frequency, and nothing else: it touches neither the filters, nor the delays, nor the polarities. It may pick, for the reference speaker's group, a different crossover from the one in place. The subwoofer alignment, computed over the full band without a crossover, still holds, and so do the filters: there is nothing to re-run at step 8. The group's previews, on the other hand, become stale and need regenerating.
The case of an earlier session. An alignment applied before the full-band search had been computed at the crossover of its reference group; an imported session that carries one keeps it as it is. For that alignment only, RCH keeps the applied pair: the speaker, and its group's crossover at that moment. Selecting another speaker does not change it. As soon as that crossover changes, a strip appears above the table naming the cause — "crossover of the Find Sub Alignment reference group changed" — that speaker, its group and the two frequencies, the applied crossover and the new one, and it carries the Recalculate alignment button. Going back to the applied crossover does not remove it: only a new Find Sub Alignment does, and it replaces that alignment with a full-band one. Nothing re-runs on its own: that button asks for a confirmation, and the same warning is written to the log.
The listening comparison between node 0 and node −1 — the subwoofers delayed by half a period at the reference crossover, polarity flipped — likewise only applies to those alignments computed at a crossover. After a full-band alignment, switching to node −1 is unavailable; returning to node 0 remains possible.
As long as that strip is showing, step 9 is not cleared: you go back to step 8. When it no longer shows, finalization can begin.
The eye buttons — one per line, and the one in the Preview column header that runs them all — build the predicted speaker-plus-subwoofer responses before transfer. This is the best way to detect a poor handoff, a polarity disagreement, or a wrong crossover choice.
Every preview shows up in the table as a first-class row, right below the speaker
row it simulates, with its own response thumbnail: it stays visible even when
Display only Selected is ticked. The first column carries a Preview label instead
of a position, and the row states when it was created. A Stale badge appears as
soon as the group's current crossover is no longer the one the preview simulated —
that frequency is part of its title, as X@80Hz.
Two accepted limitations. A change of FILTERS alone stales nothing: no signal expresses it, so regenerate the preview after recomputing filters. And the smoothing of a preview measurement is frozen at creation time (Psy): it does not follow RCH's current smoothing setting. Finally, generating a preview REPLACES the previous one for the same channel at the same position — previews no longer pile up, and only a measurement whose title is recognised as a preview is removed, never one of your own measurements. A replaced preview cannot be recovered: save an .mdat before running them all again (see the two banks).
The polarity check — the ⟳ button in the Invert column header — inverts neither speaker nor subwoofer, no more than the recompute button on each line does: it measures which polarity would blend this group best with the subwoofers, and reports it to you. Polarity itself is decided once, globally, on the subwoofer, when Find Sub Alignment runs (step 8), over the full band. Flipping the subwoofer affects every channel at once, so the check only reports a preference on clear evidence. That evidence is the height of the noise threshold — at a single position the deviation between the two polarities at the handoff must reach 2.69 dB. Reach, not exceed: exactly 2.69 dB is already enough. That threshold is the only judge of the report, not a second criterion holding a veto.
What that button measures is visible in the Polarity report card, right above the table: it shows the last run channel by channel, with the two figures the algorithm compares side by side — the summation gain at the handoff under the polarity in force, then the same gain with the subwoofer inverted.
The detailed reading of those two figures, the rule that decides when a row stops holding and the note on the handoff are in The polarity report and the previews, in detail.
Note: if a speaker genuinely needs a polarity inversion to blend correctly, it often reveals a wiring mistake (+ and - swapped). Check it physically: that is the right fix, not a DSP patch. Inverting a whole channel to compensate for a narrow band would fix the subwoofer handoff at the cost of that channel's polarity relative to every other channel.
If the previews look good on the average but less convincing at the main listening position, repeat the check while also keeping position 1 when creating averages.
This is where, and nowhere before, the LPF for LFE is decided: once the crossovers are settled.
At this step, AUTO only runs Find Best LFE Low-Pass. It saves no bank and transfers nothing to the receiver: those two gestures stay yours, once you have checked the result.
For the second bank, change only the target curve after saving the first one. AUTO then recognizes the second bank: it recomputes the speaker filters (step 5), equalizes the subwoofer sum without touching their level (step 7) and only regenerates the previews (step 9). The receiver keeps a single setup for both banks: the AUTO buttons that would change it, at steps 2, 3, 4, 6, 8 and 10, refuse without writing anything. To calibrate again, first use Clear banks.
The AVR settings card, which carries Speaker Preset, LFE low-pass filter and Subwoofer Output, stays greyed out until the bridge is connected and the receiver registered. Those three settings are not Audyssey filters: they live in the receiver, and without the bridge RCH can neither read nor write them. On an imported session the banner shows Offline calibration, all the acoustic work stays available, but that card and the transfer do not. The LPF for LFE has no lock of its own: it comes back with the chain.
Two banks coexist, and they do not serve the same purpose. Reference holds the complete calibration: speaker filters, subwoofer alignment, handoff. Flat holds the same installation without the response correction. Both are kept in the session; only one is active in the receiver at a time, and switching between them is how you hear what the correction actually does. A bank whose content was lost — browser save too heavy, incomplete file — reads “unknown” with its cause: Forget this bank empties it without touching the other one, Save current filters replaces it. A save from a version that had no banks yet simply reads them empty.
The receiver holds a single set of delays, levels and crossovers for both banks. RCH therefore refuses to save the second one if anything other than the filters changed since the first — changing a crossover between the two would force you to start over.
The summary shown before the transfer is the last chance to spot an inconsistency: it names the receiver model, correction type, channel count and saved banks. Read it again before starting the write — once sent, the calibration replaces the one in the receiver.
Each of those gestures is detailed in Finalization and transfer.
When you ask for a speaker filter, RCH computes the parametric filters for the measurement itself: this is the AutoEQ engine. This section describes what it does, what each setting changes, and how to read its verdict. You do not need to change anything to get a good result: the defaults are a proven compromise across the reference measurements.
The principle is simple: RCH compares your measurement against the target curve and places filters where the deviation justifies it. The calculation always runs in this order.
At every step, RCH only keeps the new result if it is better than the previous one. An option you enable therefore cannot degrade what RCH keeps; at worst it changes nothing, in exchange for calculation time.
Your amplifier's limits are honoured automatically. When RCH knows the receiver — bridge connected, or session imported — it bounds the AutoEQ request to what the MultEQ section can actually realise: each type (MultEQ, XT, XT32) writes filters of a fixed length and rate, which bound how fine the correction can be in the bass. On a base MultEQ amplifier the speaker correction does not reach below roughly 250 Hz — a hardware limit, not an RCH choice — while subwoofers remain finely corrected down to 11 Hz; XT and XT32 are far less constrained. Every range or Q adjustment is written to the log, never silent. Without a receiver configuration, no bound is applied.
A measurement is not only the speaker. It holds the direct sound, and then, a few milliseconds later, everything the room sends back. In the bass those reflections are the problem to treat. Higher up they blur the reading: the measured response then describes the room as much as the speaker, and a filter computed on it corrects both without telling them apart.
Windowing removes that reflected energy where it does not belong, so the calculation only sees the direct sound. The Optimized MTW preset — the default — applies a window whose length changes with frequency: nine seconds in the bass, a fraction of a millisecond at the very top. In the bass nothing is removed: every detail of the response is kept, and the filter can work finely there. That is exactly where it should be let loose.
Removing energy means removing level: the windowed curve sits below the raw one — by 1.3 to 2.4 dB between 300 and 3,000 Hz on the reference measurement corpus, for wideband speakers, and more in a very reverberant room. If the target stayed where it is, the engine would see a shortfall that does not exist and would add boosts for nothing.
RCH does not let that offset distort the dosage. At calculation time it measures the gap between your two curves — the raw one and the windowed one — octave band by octave band, and lowers the target by the same amount. In the bass the gap is nil and the target does not move; in the treble it comes down by exactly what the room was contributing. The slope you chose reaches the result intact, and the applied offset is written to the log.
Once the reflections are removed, a dip often stands out in the upper midrange. It is not the room: most often it is the handoff between the midrange driver and the tweeter of the speaker itself. It must not be corrected — filling it does not make the speaker better, and it costs filters and gain. Windowing brings it out for that very reason: made visible, it is identified for what it is and the engine does not go chasing it.
Audyssey treats the same problem differently, with midrange compensation: a fixed dip placed in the target curve at 2 kHz. The difficulty is that every speaker has its own handoff frequency — 2 kHz does not always land on it — and a dip placed beside it swaps one problem for another. Windowing imposes no frequency: it reveals the one your speaker has, whatever it is.
This is the point that misleads most often: do not judge windowing by the shape of its curve. It only sets the conditions of the calculation — what to correct, and what to leave alone. The resulting filter is then applied to the un-windowed responses: those are the ones on which the result is read, and heard.
The trade-off is deliberate. A filter computed on the raw curve would hug the target more closely, since it would correct everything — including what should have been left alone. With windowing, the corrected response departs a little further from the target on paper, and the listening result is better. What counts is what gets corrected, not how closely the trace matches.
The engine settings are in the Settings panel, AutoEQ section, laid out in five groups from the most common to the most specialised: Filters, Boosts, Q limits, Strategies, and Expert, collapsed by default. A setting that depends on another is indented under it, and greyed out while it does nothing.
Three settings weigh just as much on the result and are not in this panel: the two boost limits and the frequency range live in Speaker Tools. Those are the ones to look at first.
| Setting | Default | Range |
|---|---|---|
| Individual Max Boost | 6 dB | 0 to 6 dB |
| Overall Max Boost | 3 dB | 0 to 6 dB |
| Frequency Range | 20 Hz – 16 kHz | 10 Hz to 20 kHz |
| Setting | Default | Range |
|---|---|---|
| Filters | 20 | 1 to 20 |
| Max cut (dB) | 25 | 0 to 30 |
| Min filter gain | 0.7 | 0.1 to 2 |
| Flatness target | 1 | 0.1 to 6 |
| Setting | Default | Range |
|---|---|---|
| No boost below (Hz) | 50 | 0 to 500 |
| Setting | Default | Range |
|---|---|---|
| Narrow filters below 200 Hz | on | — |
| Max Q below 200 Hz | 8 | 0 to 8 |
| Vary Q above 200 Hz | off | — |
| Max Q in high band | 3 | 0 to 8 |
| High band starts at (Hz) | 3000 | 200 to 16000 |
Every option in this group obeys the same rule: RCH replays the calculation another way and only keeps the new result if it is better. None of them can degrade the correction; at worst they change nothing, for some calculation time.
| Setting | Default | Depends on |
|---|---|---|
| Candidate placement | on | — |
| Modal seeding (LPC) | off | — |
| Beat-REW optimization | off | — |
| Reduce & repair | on | Beat-REW optimization |
| Critical band refinement | on | Beat-REW optimization |
| Refinement | off | — |
| Refinement iterations | 100 | Refinement |
This group is closed on startup, on purpose. The defaults are the ones RCH is validated with. Open it if you know what you are after, not to explore.
| Setting | Default | Range |
|---|---|---|
| Optimization passes | 20 | 1 to 20 |
| Gain sign lock | 0.5 | 0.1 to 2 |
| Notch exclusion | 6 | 2 to 15 |
| Overshoot penalty weight | 3 | 0 to 10 |
| Overshoot reduction threshold (dB) | 1.5 | 0.1 to 6 |
At the end of each channel, RCH writes a sentence in the log. It starts with the channel name and says whether the correction is sound, usable with reservations, or refused — with its reasons, and the gesture that answers them. Three cases:
A refusal is almost never a settings problem: it means the measurement is asking for something impossible. Look at the measurement itself (microphone position, averaging), at the target curve, or at the frequency range.
You are starting out, or you want a safe result: change nothing. The defaults are designed for that. Only set the frequency range if you have a specific reason to.
You are after the best possible quality: tick Beat-REW optimization — which also activates Reduce & repair and Critical band refinement — keep Candidate placement, and try Refinement while comparing before and after. The calculation will be noticeably longer.
You want to protect your hardware: keep No boost below at 50 Hz, and lower Individual Max Boost — down to 0 dB to work by attenuation only.
The correction sounds artificial to you: that is typically an excess of narrow filters. Set Max Q in high band around 2 or 3, lower Notch exclusion so RCH ignores more dips, and check that you are not correcting too high in frequency.
You want a simpler correction: lower Filters, raise Flatness target so only large defects are addressed, and raise Min filter gain.
Keep in mind the principle that applies to all of RCH: it is the sum of the corrected measurement and your actual chain that must follow the target. Over-correction is the most common mistake.
This section gathers everything about subwoofers: the common level RCH adds to the subwoofers, the traps of measurements made in REW, the receiver LFE low-pass, aligning the subwoofers with each other and its per-sub filters, the predicted curves, MSO, and the two special cases of ADY files and the combined sweep. The workflow itself handles subwoofers at steps 6 and 7.
When the AVR calibrates its subwoofers itself, it adds to each of them a common level that depends on how many subwoofers its menu declares. That level does not show in the impulse responses: RCH adds it to the subwoofer trims, with the value of your configuration. A recent RCH Bridge gives it; otherwise RCH computes it from what the AVR declares. This holds whatever the measurements come from: the bridge measurement assistant, an ADY file, an MQX file or a Dirac Live project. The log gives the value applied at each import.
When that level is not known for your configuration (Directional subwoofer mode, a configuration not measured yet), RCH adds nothing and the log says so. After the transfer, compare the subwoofer level with the speakers, and raise the subwoofers in the AVR menu if they play too low.
What this means for you: do not add gain to the subwoofers in the AVR menu, and do not raise the sub trim gain by hand because Align SPL puts the subwoofers below the target, except in the case above — the correction would be applied twice and the subwoofers would play too loud. Earlier versions of RCH added 10 dB to every subwoofer, whatever their number: that value only held for four subwoofers. If you had lowered your subwoofers by hand to compensate, set them back to their normal value. A calibration computed with an earlier version keeps its trims: import your measurement file again (ADY, MQX, Dirac Live project) or redo the assistant measurements to apply the level of your configuration.
If a subwoofer trim exceeds the AVR ±12 dB limit, RCH warns you: turn the gain knob on the subwoofer itself and measure again.
If your subwoofer measurements were made manually in REW through the LFE channel, the AVR LFE low-pass filter can bias the curve. In that case only, unfold the "Compensate for the AVR low-pass filter (optional)" card in Subwoofer Tools and use Compensate for filter to compensate for that effect.
If your measurements come from the measurement assistant or from the Audyssey app, this step is usually not needed.
If you measured through the AVR LFE path, using the highest LPF for LFE value during the measurement session generally makes the later compensation easier.
Additional Gain does not directly change the final AVR trim. It mainly shifts the subwoofer working curve relative to the target so the optimizer can find a better solution. Individual max boost and Overall max boost define how much positive gain the algorithm is allowed to use.
With a single subwoofer, focus on Equalize Sub and Preview Sub. Align Sub and Check sub levels are then not applicable: both compare subwoofers with each other, and RCH disables them until the session holds at least two subwoofer measurements.
From 3 subwoofers on, the Best-of-orders checkbox re-runs the sequential search on several processing orders of the subs (up to 4) and adopts an alternative order only if it does not degrade the flat-target judge beyond the measurement noise (0.05 dB, bench-calibrated threshold) — the natural order always remains a candidate, so the score never regresses. Cost: up to 4 times the search time. Below 3 subwoofers the checkbox has no object (a single effective order) and says so.
With several subwoofers, the "Also optimize filters for each subwoofer" checkbox changes what Align Sub computes. It is ticked by default, like the all-pass filters checkbox, whose filters then enter the same calculation; untick it to go back to alignment followed by one shared EQ. Instead of aligning the subs first and copying one shared EQ onto all of them afterwards, the joint optimizer searches delays, polarities, per-sub level trims and up to three per-sub filters at the same time, directly against your target curve — the same idea as an MSO project, but integrated into RCH.
Why it matters: a shared EQ applied to every sub cannot change the way the subs interfere with each other. A filter applied to one sub can remove a destructive contribution at its source, which usually tracks the target much better in the modal region.
In this mode, "Allow all-pass filters (slower)" allows the optimizer to also place one all-pass filter per sub. An all-pass rotates the phase locally without touching the level: it is the only lever that can turn a destructive contribution into a constructive one instead of muting it. It helps when a cancellation dip resists the delay, polarity and filter levers; the optimizer automatically drops all-pass filters that do not pay for themselves. If you are unsure, run the optimization once with and once without the checkbox and compare the predicted curves.
Without the joint checkbox, Align Sub keeps its historical behavior: alignment only, with the all-pass checkbox helping the alignment search when enabled.
RCH maintains a virtual subwoofer per listening position. The measurement named "LFE predicted_P1" is the predicted sum of all subs with their current settings, recomputed after each subwoofer action — for the position being worked on. After Align Sub, a second measurement named "LFE Max Sum Theo_P1" shows the theoretical ceiling: the sum you would get if every sub arrived perfectly in phase at the listening position.
What this measurement represents exactly: the LFE signal the AVR will output, with one deliberate difference — the "LFE low-pass filter" (LPF for LFE), which the AVR does apply, is not applied here, because its value is not chosen yet at this stage of the workflow. You settle it during finalization, with Find Best LFE Low-Pass.
Every sum appears in the table as a first-class row, with its response thumbnail, right under the last subwoofer of its position: it stays visible even when the Display only Selected box is ticked. The first column carries the LFE sum label instead of a position, the Channel column names the signal (LFE), the Crossover column recalls that no LFE low-pass is applied, and the row shows when it was computed. No row command applies to it and nothing is ever sent to the AVR from it: it is a computed result, not a measurement. Sums whose subwoofers are not displayed (another position) are gathered at the end of the table.
A Stale badge appears when the subwoofer base changed after this sum was computed and its position has not been recomputed since. That is the case for the positions you are not currently working on: Align Sub, Equalize Sub and the settings reset only recompute the selected position, whereas Preview Sub recomputes every one of them — after it, no sum is marked. Nothing is kept across sessions: after a reload no sum is marked until a new action touches the subwoofers.
Comparing both curves tells you how much of the theoretical maximum the current alignment actually captures. A predicted curve close to the Theo reference means the alignment job is done; a persistent gap localized in frequency usually points to an interference dip worth attacking with the joint optimization or the all-pass option.
Equalize Sub computes the shared EQ on the predicted sum and distributes it to every sub. The per-sub filters found by the joint optimization, and the all-pass filters, are preserved: the shared filters are packed into the remaining free slots on every sub. This is the recommended finishing step when the per-sub filters alone are not enough to track the target closely.
The step is safe to repeat: each pass recomputes the EQ from the optimizer baseline instead of stacking on top of the previous result, so running it twice gives the same answer.
The "Min filter gain" setting is enforced everywhere: computed filters below the threshold are discarded so near-zero corrections do not waste filter slots.
If you work with several subwoofers and MSO, RCH can export sub measurements in a format suitable for initializing an MSO project, then reimport into REW an Equalizer APO export produced by MSO.
This block appears only for multi-sub configurations and assumes that REW is connected and that measurement names are correct. The export carries the takes only: MSO works on the measured listening positions, not on averages. To export, keep the measurements when you create the averages.
Simple principle:
1. Export sub measurements from RCH
2. Work inside
MSO
3. Export from MSO in Equalizer APO format
4. Import that text file
back into RCH
The MSO export menu used for this workflow is shown below. Click the thumbnail to open the full-size image.
If you need to finish the import manually in REW, use the EQ window with the text file created for that project.
For measurements imported from an ADY file with multiple subwoofers, directional bass mode matters: without it, the file may not contain enough usable individual subwoofer measurements for a real multi-sub workflow.
Such a file still imports: RCH never refuses an import on the sole word of the file. If the measurement conditions the file declares (subwoofer mode, number of subwoofers) do not match the receiver configuration RCH has loaded, RCH writes a warning in the log and the import goes on — it is up to you to judge whether those measurements are still relevant for the receiver in place.
RCH never changes the subwoofer mode of the receiver: the transfer leaves it in whatever mode it is in. Set the mode you want from the receiver's own menu before measuring (an automatic Directional / Standard switch driven by RCH bugged the real receiver, so it was removed).
You do not have to guess what your receiver is capable of. What matters is what RCH actually received: one measurement per subwoofer, or a single one for all of them. Some receivers sweep the subwoofers one by one even in Standard mode, others only in Directional bass mode, others never — and the measurement file decides nothing. RCH tells you which case you are in.
If the note Your amplifier returns a single combined response for its subwoofers appears under the subwoofer tools, all the subwoofers were measured together as one sweep. RCH then holds a single subwoofer measurement: Align Sub and the Subwoofer level check card stay unavailable, and the calibration of that single measurement applies to every declared subwoofer. Nothing is broken and no step was missed: this matches the official behaviour. When the receiver resolved its subwoofers separately on its own, the Subwoofer level check card shows the levels it reports for each of them: they are there for information, RCH neither computes nor applies them.
One generation of receivers is the exception, and RCH stays silent on it: those that do not declare their subwoofer configuration (AVR-X4300H, AVR-X4400H, AVC-X4700H). On those, two identical sweeps of the same subwoofer, one after the other, do not return the same figures — up to 9 cm apart on a distance, and half a decibel on a level. Those are not measurements, they are noise, and RCH would rather show nothing than print a figure that would make you want to correct something. Those receivers lose nothing by it: with a recent RCH Bridge they now sweep each subwoofer for real, which is far better.
If a suggestion to switch to Directional bass mode appears next to that note, your receiver offers that mode and it did not sweep your subwoofers separately in its current one: switch it from the receiver menu, then measure again. One counterpart to know about: the mode is part of the receiver state a session is calibrated for, so a session exported in Standard is refused on import while the receiver is in Directional, and the other way round. If that suggestion does not appear, RCH did not identify a useful switch on your receiver — the note itself stays true.
The combined sweep does not change the level matching, which stays per subwoofer: when the receiver supports it, the interface offers one button per declared subwoofer before position 1, and only the subwoofer being adjusted plays. The other ways of getting one measurement per subwoofer are still open: repeat the Audyssey session in Directional bass mode, or measure each subwoofer manually in REW.
These notes are kept with your work: they come back after a page reload and they travel inside the session file, so that whoever receives a calibration knows it came from a combined sweep.
This is the opposite situation to the one above, and the two must not be confused. Here your receiver did sweep each subwoofer separately, but two of the responses it returned are identical. RCH says so under the subwoofer tools: RCH received identical responses for SW1 + SW2.
Nothing is blocked and your calibration works. What the note reports is that those two measurements most likely describe a single subwoofer: alignment between those two then has nothing real to work on, and correcting it amounts to correcting the same loudspeaker twice. That note is kept with your work, like the previous ones, because it is mostly read after the campaign, while looking at the measurements.
If your two subwoofers really are distinct — different placements, different models — then your receiver did not honour the request, and that is a case nobody has observed yet. Export your session and report it: that file carries the channel plan, the receiver state and the measurements, which is exactly what is needed to study the case.
What follows describes what the Polarity report card publishes, how its two figures are read, and when a row stops holding. Step 9 of the walkthrough says what to click and what to look at; this section says what the numbers mean.
That figure says how much the speaker and the subwoofers add up in the zone where they share the work — and, when it is negative, how much they cancel each other there instead. The bigger it is, the better the join: the maximum is +3 dB, reached when the two paths arrive in perfect agreement. Do not confuse that gain with the Junction deviation (dB) column in the table: it measures something else and reads the other way round, the smaller value being the better one there.
Both figures are measured at the crossover in force on that channel, shown at the head of the row — the same point for both polarities, which is what makes them comparable; and since the crossover changes from one channel to the next, two rows cannot be compared without it. The bigger of the two gains is the better join: that is the one RCH keeps, and it shows in green. Then comes the decision itself (conforming, would prefer this speaker inverted, flagged but not decided). When the gap between the two figures does not clear the measurement noise, neither turns green and both carry a "≈": RCH reports without deciding. The noise threshold itself, along with the measured gap, is written to the log.
This card offers no button: there is nothing to apply channel by channel. Collapsed, its header sums up how many channels were measured and the subwoofer verdict. That report is the result of a run, not a state of your session: it is not saved and it reads "not run" again after a reload. The same table is written to the log, between the operation's rules: that copy survives a reload and travels with the log export.
During a session, staleness is TARGETED, row by row: each row only holds for the state of the settings that built ITS summation — its speaker, its group's crossover, the subwoofers of its position, and the filters of all of those. As soon as one of them moves, that row is emptied and reads "not evaluated"; the others keep their verdict, which is still true. Run the check again to fill it back in. Note that a change on the SUBWOOFER side (level, inversion, delay, EQ, optimizer) empties every row, and that is not overzealous — the card only shows one position at a time, and all of its rows judge the SAME subwoofer sum. The card header then states the verdict of the rows that are still valid, and how many are not; when no row is evaluated any more, it says so in as many words — not to be confused with "criterion unavailable", which flags a measurement the tool could not judge and calls for a different look entirely.
Note on the handoff: the AVR's bass management is asymmetric — it applies an LR24 low-pass to the subwoofer but only a BW12 high-pass to the speaker, betting that the speaker's natural roll-off completes that BW12 — yet the EQ has just flattened that roll-off, which would push the sum about 2 dB above the target around the crossover. RCH corrects this by design: a second BW12 is baked into the transferred filters of every bass-managed speaker, so the real junction becomes a symmetric LR24|LR24 pair, in phase at every frequency. Previews and Find Best LFE Low-Pass simulate that complete chain: what the preview shows is what will come out of the AVR. The Check time alignment measurement works on the raw equalized curves — the LR24|LR24 pair being in phase at every frequency, it does not invalidate that measurement. Find Sub Alignment also works on the equalized curves, without any handoff filter: its best full-band sum does not guarantee the best sum once crossed over, and the previews are what show it. Find Best Crossover, for its part, simulates that LR24|LR24 handoff for every candidate frequency: it is precisely because the handoff is neutral that it can simply compare the resulting response to the target. Point to watch in the previews: below the crossover, where the speaker rolls off naturally and the EQ could not compensate, the second BW12 slightly deepens the dips — that is the region where the subs dominate, and a well-chosen crossover (Find Best Crossover) avoids the unfavourable pairings.
The acoustic assessment reads your measurements and tells you what they are worth: one grade per measurement, the detail criterion by criterion, and a list of things to do, sorted by urgency. It changes nothing — not your measurements, not your filters, not the project open in REW. It opens with one click on the response thumbnail of a measurement, in the Response column of the table: the window that shows up starts with that very thumbnail in full size, then the criteria one by one. The two buttons in the header of that same column — Assess the measurements and Compare before / after — grade every measurement of the position on display in one go. The table itself carries neither a column nor a grade: there is nothing else to open.
Assess the measurements, like opening a thumbnail, reads the measurement as it stands: for a speaker, the measurement itself, before any correction — the position average if you made one; for the subwoofers, the "LFE predicted" projection of their position, on its own, filters included. Every line gets a grade from A to D, the detail of its criteria — eighteen for a speaker, five for the LFE line — and the actions that follow from them.
The grade is read in the assessment window, and nowhere else: click the thumbnail of the line to open it. You do not have to grade everything at once — the two buttons in the Response header handle every measurement of the position on display, one opening grades just one of them, in about a second.
Subwoofers are the exception. The thumbnail of a subwoofer only opens its response view — its measurement, its corrected curve and the target — with no grade and no criterion: the impulse response of a single subwoofer is not what the amplifier will put out. The assessment of the subwoofers is the one of their projection, and it is read on the "LFE predicted" line of their position: open its thumbnail. This assessment opens like that of a measurement assessed on its own — one letter, no switch, no verdict — and both Assess the measurements and Compare before / after grade that line. A position without a projection has no LFE line.
The LFE line carries five criteria only: deviation from target (20-200 Hz), uncorrectable dips, low-frequency extension, delay in the bass and signal-to-noise ratio. The others read a band a subwoofer does not reproduce (the tilt over 100 Hz-10 kHz, the midrange, the reflections over 200 Hz-5 kHz), set the bass against the midrange of the same measurement (bass decay, modal tail), or judge the crossover of a speaker: they do not show on the LFE line, and its grade counts the five alone.
Compare before / after gives each line that has a preview its after. Each preview — the "final …" measurements created from the Preview column at step 9 — is matched to the measurement it came from by its title; the LFE line, for its part, stays graded on its own (see Comparing before and after). Both sides go through the same calculation, the same window and the same smoothing; the target curve is read once, on the before, and reused as is on the after. Without that, you would be comparing two different readings rather than two states of the system.
What the assessment is not: neither a step of the walkthrough nor a lock. Nothing is disabled because of it, no result is refused, and it writes nothing into REW. It is not kept either: reload the page and every assessment is computed again the first time it is opened.
Every quantity is computed by RCH from the impulse response read in REW. REW is used to check those computations, it is not their source: the reading is therefore the same before and after, and the same from one run to the next.
The response — what the curve shows:
The time — what the room adds after the direct sound:
The phase — what arrives when:
The measurement itself, and the crossover:
A criterion can stay ungraded without anything being wrong: the band it needs falls outside the usable band of the channel, the analysis window is too short, there are fewer than three positions, the channel has no mirror, the target could not be read. Every time, the assessment writes the reason out in full. And when a criterion simply does not apply — the extension of a speaker crossed to the subwoofers, the noise floor of a receiver measurement — it does not even count in the tally of evaluated criteria.
The grade is a letter, from A (the best) to D, set in a square of its own colour, to the code of a nutrition label: dark green for A, green for B, yellow for C, orange for D. The colour and the letter say the same thing, and no word goes with them. Without a grade, the square stays empty, outlined with a grey dashed line — the absence of a grade, never a bad one. Hovering the badge, and the screen reader, do spell the scale out: "Grade B, on a scale from A to D".
There is one grade per measurement and never a system average: a single figure would hide the channel that goes wrong.
The grade is built like this: each letter is worth points — A four, B three, C two, D one. Criteria are first averaged within their family, then the families are weighted: the response for 40 %, the time for 30 %, the phase for 15 %, the crossover for 15 %. A family with no evaluated criterion does not weigh, and its weight is shared among the ones left. Two safeguards then apply. The signal-to-noise ratio caps the grade at its own. And the worst of the three critical criteria — deviation from target, early reflections, delay at the crossover — caps the grade one letter above its own: a D on any of them holds the whole line at C at best. The assessment names the criterion that capped.
When fewer than 60 % of the criteria of a measurement could be evaluated, the grade is still shown but the assessment says how many criteria carry it: a grade resting on four criteria is not a grade resting on sixteen.
A click on the response thumbnail of a line — or the Enter key once the thumbnail is reached from the keyboard — opens the assessment of that measurement: a window that shows one card at a time, in full size.
The first card is the thumbnail in full size. It carries the measured response at full resolution, the response corrected by the filters currently loaded in REW — with no bass management — and the target curve, on graduated axes, in the same 35 dB window and on the same centre as the thumbnail in the table. It is there to get your bearings before you go into the detail. It is not graded: no letter goes with it. And it is the only card of the window that is read from REW for display.
Then come the criteria, one per card: the name of the criterion, its grade, its value spelt out, and a chart of what it looked at — the part of the chart that gave the grade is bounded in purple and named, the rest is veiled. The chart shows what the criterion read; the badge says what it makes of it.
There are four sorts of charts, according to what the criterion measures: the frequency response (frequencies across, decibels up), the group delay (frequencies across, milliseconds up), one bar per third-octave band (decay, modal tail, clarity) and time (the envelope of the impulse response, in milliseconds or in seconds). Under each chart, a one-line legend names only the curves actually drawn — the measurement (the projection, on the LFE line), the target, the paired channel, the positions, the after — and, in comparison, the entry of the original measurement, which shows or hides it with a click. The scale of the criterion is spelt out with its source: those are the bounds that gave that very grade, not a generic scale.
The charts of the criteria come from the same numbers as the grade, never from the curve displayed in REW: same smoothing, same target, same window, same level (see Deviation from target). That smoothing is the one of the assessment, a perceptual one: a forty-eighth of an octave below 100 Hz, to keep the modal detail, widening to a third of an octave above 10 kHz, where the ear integrates over broad bands. Only one chart is smoothed to be read, the group delay: at a sixth of an octave, weighted by energy as the grade is weighted by band; the legend says so. The subtitle shows the position, the role of the measurement — raw measurement, average of the positions, preview, LFE projection — and the time of the reading. Below the chart and its legend, a collapsed block opens the entry of the criterion — what it reads and what earns each letter — without leaving the window. Under it comes the summary of the criterion: the method in one sentence, the bounds of the scale, then the actions this criterion calls for, each of them linking to the page that explains it.
Explore the criteria. The chevrons sit on the flanks of the chart, halfway up, see-through. Above the name of the criterion, the row of markers groups the criteria by family, on a single line: each criterion is a segment in its grade's colour, outlined in dots without a grade. The segment of the open card is taller, and the name of its family turns bold. A curve icon opens the row: that is the response view. A click opens its card and keeps focus. Left and right arrows change card, Home returns to the response view, End goes to the last criterion and Esc closes the window.
The window keeps a single column. The button next to the cross enlarges it; the chart fits the space available while keeping its proportions. The bounds of the scale are read in the summary of the criterion; the source of each bound — the standard cited or the dated corpus — is read at the foot of the “Understanding this criterion and its grade” block, which explains what the criterion reads and what earns each letter. Expanding this block moves nothing: its header stays where it is, the chart keeps its size, and the documentation takes the place of the summary, with its own scrolling. Without an automatic recommendation, the measured value and scale remain the reference.
The window fits without scrolling in a browser window at least 768 pixels high, which any desktop screen offers. Lower than that — a laptop whose bars eat into the room — the middle part scrolls on the busiest cards, while the title and the controls at the foot stay where they are. RCH is not made for a phone: it needs the bridge and REW on the same machine.
Recompute, at the foot of the window, reads the measurement again and redoes the assessment without closing it: the card on display stays the same, only the figures and the letters are refreshed. Switching to the comparison or back to the raw measurement also keeps the current criterion when it exists in both views. That is the thing to do after loading other filters into REW — a change RCH does not see go by, and which therefore makes nothing stale on its own. The button is dimmed while a calculation runs and as long as RCH cannot read from REW; its tooltip then says what is missing.
The states: a measurement that has not been graded yet is computed when its window opens, and the window says it is reading meanwhile; if the reading fails, it says why. A measurement whose readings have changed is computed again the same way, on opening. When RCH cannot read again — REW closed, or a workflow running — the assessment already computed is shown as it stands, dated, and the window says it is no longer up to date: Recompute reads it again as soon as REW answers. Nothing survives a reload of the page: every assessment is computed again the first time it is opened. A criterion that is not evaluated, for its part, keeps its frame, veiled as a whole, and states its reason under its name, in place of the value — the band it needs falls outside the measurement, the window is too short, positions are missing.
When a criterion is not evaluated. The reason is read under the name of the criterion, the frame stays veiled and writes nothing, and a sentence of the summary says what to do. Three cases come up often. An Audyssey archive does not keep the silence that precedes the impulse and stops shortly after the arrival: the noise floor cannot be measured there and the decay cannot be read — measure that channel with REW and import the measurement. A speaker crossed to the subwoofers no longer owns its bass: extension and bass are judged on the LFE line of the position, and there is nothing to do. A channel with no mirror channel — the centre — is compared with no pair: nothing to do there either. When the mirror channel has nothing comparable at the same position, symmetry waits for it: measure that channel, or create its preview at that position. And the grade does not suffer for it: a criterion that is not evaluated is left out, and its weight goes to the others of its family.
Eighteen criteria, spread over five families: the response, the time, the phase, the measurement itself and the crossover. Each entry says what the criterion reads, how its value is obtained, what earns each letter and with which bounds, when it stays ungraded, and what a low grade calls for. The very same entries are read without leaving the assessment window: under the actions, the collapsed block Understanding this criterion and its grade unfolds the text of the criterion on display. The entry that opens the list is not a criterion: it is the card the window opens on.
The first card of the window is the thumbnail of the line, in full size. It carries the measured response at full resolution, at the smoothing RCH applies, the response corrected by the filters currently loaded in REW — with no bass management — and the target curve, in the same 35 dB window and on the same centre as the thumbnail: what you see small in the table, you find again here in full size, in the same proportions. Those three curves are read from REW for display, and for display only; the criteria that follow are read on the impulse response instead. This card therefore gets no letter.
Measuring a band of the response
Below the measured response, the Band metrics panel reads the measured curve, at the view's smoothing, between two frequencies: the slope of its trend line in dB per octave, its deviation from the line above and below, the RMS deviation, the share of the band that stays within the chosen tolerance, the maximum, the minimum and the average. The line, its two dashed tolerance bounds and the minimum and maximum points are drawn on the graph.
The band can also be set with the mouse, on the graph: drag an edge of the band to move it, or drag elsewhere to draw a new band. The edge's frequency shows during the gesture; on release, the From and To fields take the edges, to three significant figures. The fields remain the keyboard path.
The measurement has a single method, the most accurate one: each octave counts the same, the band edges are read exactly, and the average is taken in energy. The slope therefore does not depend on how fine the measurement is, and the average is the level a sound level meter would read on pink noise limited to the band. Tick Relative to the target: the values become gaps to the target, and the graph then shows the band only.
On an average, a preview or a predicted sum, which REW holds on a logarithmic grid, the measurement gives the slope REW reads, within one or two hundredths of a dB per octave, its minimum, its maximum and, below 1 dB per octave, the same deviation from the line. It differs on purpose on three points. Its average counts each octave the same; REW's counts each hertz: from 100 Hz to 10 kHz, the octave from 5 to 10 kHz makes up half of it. On a response that falls towards the treble, REW therefore reads lower, by a few tenths of a decibel to nearly two. An extreme at the edge of the band is read at the edge; REW takes it on the nearest measured point, sometimes just outside the band. Above 1 dB per octave, REW's deviation from the line moves away from RCH's, measured in dB at equal frequency. On a raw measurement, REW counts each measured point, and the treble holds nearly all of them: its slope mostly follows the top of the band, whereas RCH's does not depend on how fine the measurement is.
A frequency can be written “1500”, “1.5k”, “1.5 kHz” or “10,000”; “10.000” and “10 000” also mean 10,000 Hz. An empty field takes the value shown in grey. The “Calculation steps” block tells which points were measured: grid type, points used, curve smoothing, band edges and method.
It compares the smoothed frequency response of the channel with the target curve, over the usable band of the measurement, and sums up the difference as a root-mean-square distance per band. The midrange, from 200 Hz to 2 kHz, carries the grade of a speaker; the bass, from 20 to 200 Hz, carries the one of the LFE line.
The charts of the criteria show the curve at its own level, the one of the measurement: nothing shifts it towards the target. A measurement that plays 3 dB above its target is drawn 3 dB above it, and those 3 dB count in the graded deviation; the first card shows it as REW reads it, at the same level. The zone, like the value, names the band actually graded, clipped to the usable band. When the level of a measurement is unavailable, RCH does not draw its response and does not grade its deviation: it says so.
For a speaker:
For the LFE line the bounds are 3, 5 and 8 dB: the modal range is rougher than the midrange.
Without a readable target curve in REW, or with too few points left in the usable band, the criterion stays ungraded. A low grade is taken up first on the target, then on the equalisation; a deviation concentrated on one narrow dip is not filled in at all.
It looks, on the response of the channel, for dips that are both narrow and deep: more than 10 dB below the local trend of the curve, with a Q above 10. The graded value is their count. A wide dip does get corrected and does not count.
The search runs within the band where the measurement responds, which the window shows as the graded zone. That band is read on an envelope of the curve — a sliding median over one and a half octaves, 6 dB below the reference level — that ignores dips: a deep dip in the middle of the bass does not cut it. It is not the −6 and −10 dB extension of the neighbouring criterion. Nothing is searched below 20 Hz or above 20 kHz, and a dip cut by the edge of the band is not counted. When the band cannot be determined, the criterion stays ungraded.
A dip is counted unless it is known to be missing at one measured position: it then only describes that seat. With the responses of every position, the window says "present at the N measured positions". Without them — always the case for the corrected preview —, or when a position is unreadable or does not cover its frequency, the dip is counted and the window says "presence at the other positions not assessed". The look of a dip alone does not tell whether it comes from a room mode or from an interference with a wall: how it is shared across the positions is what points to the remedy.
Equalisation does not fill a dip of that kind: it moves when the seat or the speaker moves. See A deep bass dip, at every position and A wide dip on one speaker only.
It reads the bottom of the response and takes the frequency at which it has fallen 10 dB below its reference level: how low the channel really reaches, with the room included in the measurement.
For a speaker measured on its own:
For the subwoofers the bounds are 20, 25 and 32 Hz.
A speaker crossed to the subwoofers no longer owns its bass: the criterion does not apply, and it does not even count in the tally of evaluated criteria. It also stays ungraded when the usable band cannot be determined. A short extension is not corrected by equalisation; it is dealt with at the crossover, by handing the bass over to the subwoofers earlier.
It fits a straight line on the response, within the part of 100 Hz to 10 kHz the measurement covers, does the same on the target curve within that very band, and compares the two slopes. The slope of the target is the target's alone: it is the band that depends on the measurement, and the chart shows it while the value names it. The graded value is the difference between the two slopes, in decibels per octave, without its sign: a response that is too dark and one that is too bright depart just as much from what was asked.
Without a readable target curve, or with too few points left in the usable band, the criterion stays ungraded. It does not show on the LFE line: a subwoofer does not reproduce the band from 100 Hz to 10 kHz.
A tilt difference is taken up on the target first — a dull room calls for a less steep target — then on the wideband equalisation. It is not a fault of the room but a disagreement between what it gives and what is asked of it.
It sets the response of the channel against the one of its mirror — front left against front right, left surround against right surround. Both channels are compared at their own level: two speakers with identical response shapes, one playing 3 dB louder, are 3 dB apart, and their curves are drawn 3 dB from each other. The relative level, given separately, says which one plays louder. The root-mean-square distance is read in the midrange.
A channel with no mirror — the centre — stays ungraded, and that does not weigh on its grade.
A pair that drifts apart usually owes it to its surroundings: one speaker near a corner, the other in free space. See A wide dip on one speaker only.
It compares the responses of one channel from one listening position to the next and reads their spread in the midrange. It is the only quantity of the assessment that cannot be read on a single measurement: at least three positions are needed, and the criterion names in passing the one that clearly stands apart.
With fewer than three positions the criterion stays ungraded. It only exists on the side of the original measurement: a preview is computed from one curve, it has no positions to compare.
A wide spread is expected across a three-seat sofa; one position on its own is rather a measurement to take again. See A measurement too noisy.
It reads, third-octave band by third-octave band from 500 Hz to 2 kHz, how long the sound takes to die away once it stops, and keeps the median of the usable bands. At least three are needed. It is the only time quantity of the assessment that is compared with a listening-room standard, and only in that band.
The grade is read inside a window, with a floor and a ceiling:
A room that is too dull falls out from below, a room that is too reverberant from above. The criterion stays ungraded when the usable band does not cover 500 Hz-2 kHz, when the window is too short, or when fewer than three bands can be read. See The room rings for too long.
It reads the decay of the bands from 63 to 160 Hz and sets it against the midrange decay. The graded value is that ratio, not a time: in the bass an absolute duration means nothing, for want of a standard to compare it with. Above 1, the bass rings on longer than the rest of the spectrum.
Without a usable bass band, or without a readable midrange decay, which serves as its reference, the criterion is not evaluated, and its reason names what is missing: the bass if it cannot be read, otherwise the midrange. It does not show on the LFE line: it sets the bass against the midrange of the same measurement, which a subwoofer does not reproduce.
A bass that lingers is a room mode, not an equalisation fault: lowering the level barely shortens the tail. See A bass peak that rings on after the sound.
The same reading as the bass decay, but band by band between 25 and 200 Hz, and on the most lingering one of them. The criterion therefore says which band rings on, and by how much: this is the figure that replaces reading a spectrogram.
When no bass band can be read — a window that is too short, a measurement reduced to a single impulse — or when the midrange decay, its reference, cannot be read, the criterion is not evaluated, and its reason names what is missing: the bass if it cannot be read, otherwise the midrange. It does not show on the LFE line: it sets the bass against the midrange of the same measurement, which a subwoofer does not reproduce.
The band that is named points at a room mode. A subwoofer moved, a tuned absorber or a seat brought forward act on it; equalisation only lowers its level. See A bass peak that rings on after the sound.
It sets the energy of the first fifty milliseconds against everything that follows, third-octave band by third-octave band from 100 Hz to 2 kHz, then averages the bands it keeps. A band is kept only if the signal there rises at least 30 dB above the noise floor, failing which the figure would describe the noise.
When no band is left with enough dynamic range, the criterion declares itself unreliable rather than giving a figure.
A low clarity is taken up on what arrives early: the first reflections, the furniture around the speaker, the distance to the wall behind the listener. It is not corrected by a filter.
It walks the envelope of the impulse response between 200 Hz and 5 kHz and picks up the distinct returns that follow the arrival, from half a millisecond to twenty milliseconds. The grade comes from the strongest one within the first fifteen milliseconds, read below the direct sound. What arrives before 1.5 ms is reported apart: it comes from the speaker itself, not from a wall.
When the usable band of the measurement does not cover 200 Hz-5 kHz — a speaker crossed over high — the criterion is not graded, and it says so: the band is not covered. A measurement whose window is too short stays ungraded as well.
The delay of the return gives the distance of the wall at fault. See A reflection that is too strong right after the direct sound.
It reads the excess group delay of the direct sound — over the first eight milliseconds of the impulse response, so that the room does not enter the count — third-octave band by third-octave band from 500 Hz to 8 kHz. In each band it takes the average delay, weighted by energy and keeping its sign, then sets its absolute value against the threshold above which the ear hears that delay, a threshold that depends on frequency. A negative delay counts as much as a positive delay of the same size. The grade comes from the worst band, as a ratio to the threshold rather than in milliseconds.
The window shows the threshold on both sides of zero, over the graded bands only, and the average of each band; the worst one is circled. The background curve is smoothed over a sixth of an octave to be read, whereas the grade averages each third-octave band: a narrow spike stays visible on the curve without necessarily weighing on the average, which is why the letter does not always follow the peak the eye sees.
It only reads the third-octave bands lying entirely within the band where the measurement responds, and it needs the one at 1 kHz; a measurement that does not cover it leaves the criterion ungraded, and it says so: the band is not covered. An undetermined response band, an impulse response reduced to one impulse, or too few usable points, leave it ungraded as well.
In that band, an excess comes from the filters or from the alignment, seldom from the room.
It reads the same excess group delay between 20 and 300 Hz, but on the whole impulse response: in the bass the delay is in the room, not in the first arrival. It only reads the part of that band where the measurement responds: a speaker that only goes down to 150 Hz is read from 150 to 300 Hz, and a speaker that starts above 300 Hz is not graded. The graded value is the average over the band read, weighted by energy, keeping its sign, then taken in absolute value.
The scale is not in milliseconds: it is read as fractions of one period at the frequency being judged.
A floor of 9.5 ms applies: below it there is nothing left to correct, and a tighter bound would only describe itself. The scale is read at 100 Hz, where one period lasts 10 ms: a quarter and a half of a period both fall under the floor, so A and B are both 9.5 ms and C is 10 ms; there is therefore no B at 100 Hz. The window draws the average as a segment over the band read, and the bounds on both sides of zero. An undetermined response band, a degenerate impulse response, or one with no usable point in the band, leaves the criterion ungraded.
An excess in the bass follows the room modes and the placement of the subwoofers; it is taken up there, not at the filter.
It reads the same excess around the crossover frequency, from one octave below to one octave above, on the whole impulse response, and only over the part of that band where the measurement responds. The graded value is the average over the band read, keeping its sign, then in absolute value. The reading at the crossover point itself is given alongside, for information only: it is a narrow, unstable spike that the assessment does not grade, and it only exists when the crossover lies within the band where the measurement responds.
As in the bass, the scale is read as fractions of one period:
The same floor of 9.5 ms applies; above about 105 Hz a whole period falls under the floor and all three bounds are 9.5 ms. The criterion only exists on the "after" side: before correction there is no crossover to judge yet. It stays ungraded when the channel has no crossover frequency — a speaker that runs full range —, when the response band is undetermined, or when it does not meet the crossover band.
It is one of the three criteria that cap the grade. See Find the best crossover automatically.
It reads the sign of the first arrival of the impulse response: positive when the driver first moves towards the listener, negative when it moves away.
This criterion gets no letter, and that is deliberate: the polarity decision is a global one, it is taken once for the whole system in the polarity report, and a measurement taken on its own cannot settle it. The assessment therefore writes the finding down without grading it, and that finding does not enter the tally of evaluated criteria.
The sign is read between 200 Hz and 5 kHz: when the arrival does not stand out enough there, it is not read, and the criterion is marked not reliable — still a finding, never a grade.
A negative polarity found on several channels is dealt with in the polarity report, never channel by channel. See The polarity report and the previews.
It reads the silence that precedes the impulse, as the measurement keeps it — 10 ms at least, taken 5 ms before the arrival so as to catch nothing of the attack — and compares that noise floor with the peak of the measurement. The frame REW adds around an imported measurement is not silence: it is set aside.
It does not weigh in the average: it caps the grade at its own. Without silence ahead of the impulse — a receiver archive — it stays ungraded, and then caps nothing. A preview carries over the ratio of its original measurement: a computed signal has no noise floor of its own.
On the LFE line, it is read on the sum of the measurements of its subwoofers, and the chart shows that sum: a projection is computed, it has no noise floor of its own. The noise is read there at least 50 ms before the arrival, not 5 ms: what comes closer to it is the foot of the impulse, not noise. Measurements taken by the receiver do not keep that silence: the criterion then says how many milliseconds they keep, and the grade is not capped. To have it graded, measure the subwoofers with REW and import the measurements.
A noisy measurement is taken again, louder, or later. See A measurement too noisy.
It computes nothing: it reads, as it stands and with its sign, the summation gain of the polarity report — by how much the speaker and the subwoofers add up where they share the work, at the crossover where the report judged them. Zero decibel means they add up without loss; a positive figure, that they gain by playing together; a negative figure, that part of the bass is lost in the handover.
That gain is not the deviation from the target: it compares the sum of both paths with their separate energies, without looking at the target. A sum can stay below the target with a positive gain. The window therefore draws it apart, as a bar from zero at the report frequency, with its bounds, never on the response curve. The verdict alongside is the polarity report's: "conforming" says the current polarity is the right one, not that the response is above the target.
When the speaker and the subwoofers play together, their sounds can reinforce or cancel each other. This criterion measures that effect where they share the frequencies: it compares the level of their sum with the reference of their two energies added, weighted over their overlap. As an illustration, unrelated to your measurements: two paths of 70 dB each have a 73 dB reference; if their actual sum is 74.7 dB, the gain is +1.7 dB.
The criterion only exists on the "after" side, and it only grades a preview when the report was computed at that preview's crossover. As long as the polarity report has not been run, is out of date or could not judge this crossover, or judged it at another crossover, the criterion says so and stays ungraded.
A loss at the crossover comes from the polarity, from the delay between the speaker and the subwoofers, or from the crossover frequency: those are the three settings to take up before any filter.
It reads, in the polarity report as well, the deviation between the sum of the speaker and the subwoofers and the target curve, across the whole crossover band, from 20 to 320 Hz. Level counts: a bump or a dip at the junction shows up here, and so does a sum that sits below the target across the whole band.
Like the summation gain, it only exists on the "after" side and waits for the polarity report to have been run.
A shape that departs from the target is taken up through the crossover frequency and the delay of the subwoofers; equalising the sum comes afterwards, not before.
Matching is done by title: a preview carries the title of the measurement it came from, and that measurement becomes its "before". There is no guesswork behind it — if the title matches no measurement, the line is graded on its own and the assessment says so.
The LFE line is not compared: its projection is graded on its own. The sum of the aligned subwoofers — delays, polarities and levels set — does not serve as its before, for it has none of their filters, neither the equalisation of each subwoofer nor the all-pass the optimizer may add to it: with several subwoofers, that state never exists.
Both sides go through the same calculation, the same window and the same smoothing. The smoothed curve of a preview is never set against the raw curve of its source, and the target is the one used on the before.
A difference is only called "better" or "worse" beyond a comparison threshold. It is a convention, based on the variability between your listening positions — their spread, level differences included, when you have at least three of them —, otherwise on a spread measured on a large body of measurements. The assessment says which of the two it used. The spread between seats does not measure the uncertainty of a difference read at the same seat: this threshold is a safeguard, not a statistical confidence. Up to it, the difference may well be real, but the assessment does not claim it: the verdict is "equivalent". The differences stay within the comparison thresholds in use.
A letter is not a change. Each side gets its letter on its own scale, whereas the verdict sets the difference between both values against the comparison threshold: a letter can change without the difference crossing that threshold — two dips counted after correction against none before, for a threshold of two —, and the window then says so in so many words, under the chart. Likewise, each side is read over the band where its own measurement responds; when the two bands shown differ, the window names them, and the two values are not two readings of the same range.
The verdicts of a criterion are therefore: better, equivalent, worse, or not comparable — a criterion that can be measured on one side only is never called equivalent. For a whole line, "mixed" is added, when criteria go both ways. A critical criterion that goes backwards takes the whole line backwards with it.
The comparison is read in the assessment window. Opened from the thumbnail of a preview, the window is already in comparison; opened from the raw measurement, Compare with the preview only shows up when a preview of that measurement exists, and Back to the raw measurement returns to the earlier reading. Both buttons sit in the same place: you compare by switching, without moving the mouse. Switching moves nothing in the window — only the chart, the values and the grades change — and both sides have the same cards: the three criteria that only exist once the crossover is in place also show on the raw measurement side, in a veiled frame, with the action that points to the preview.
The banner holds on one line. In comparison, it carries two badges, before and after, the number of common criteria they are read on, the verdict of the line and its three counts: an arrow up for the criteria that are better, an arrow down for those that are worse, an equals sign for those that are equivalent. When a criterion caps the grade, the banner names it at the end of the line, with its letter, and a click leads to its card.
On the first card, “Compare with the preview” displays the preview response, with its processing already applied, and returning to the raw measurement brings back its original curves. Each is read from REW only once per opening of the window: switching back reads nothing again. The title and legend identify the displayed side.
Each criterion in turn shows its value and its two badges, then, on a line of its own, its verdict, the threshold used and where that threshold comes from. A criterion measured on one side only says which one there; when the two sides do not compare, the line stays empty — the empty square already says so. Both curves are read on the same chart, the before set back under the after. That original measurement is only drawn on demand, from its legend entry, Before (raw measurement): a click shows or hides it, and its name stays struck through while it is hidden. It is hidden by default so as not to weigh the reading down, and the choice is kept until the page is reloaded. The entry only appears on the charts that can draw it.
A line goes stale as soon as what it read has changed. The "before" side goes stale if its measurement is written over; the "after" side goes stale as well when the crossover, the filter banks, the subwoofer baseline or the previews change. Reopening its thumbnail is then enough: the assessment is redone on opening. When RCH cannot read again, the window shows the dated assessment and says so; Recompute reads it again as soon as REW answers. The other lines do not move.
RCH knows neither the dimensions of your room, nor its materials, nor where the microphone and the speakers are. All it sees is impulse responses. The distances that appear in the suggested actions — how far to move a seat, how far away the suspected wall is — are therefore figures deduced from a frequency or from a delay. They tell you where to look; they do not replace a tape measure.
Below roughly 200 Hz, a decay time is not the reverberation of the room: it is how long one room mode keeps ringing. The assessment uses it for the bass-to-midrange ratio and for the modal tail, never to compare it with a standard. It also refuses to read a decay when what is ringing is the analysis filter rather than the room.
A measurement that is too short — less than 0.35 s of window — does not allow the long reading of the decay: the assessment falls back on the short one and says so. A measurement reduced to a single impulse, as some receiver files are, is declared unusable: what is read from it — time, phase, and the response the assessment computes from it — is left unevaluated, nothing is graded D. In the same way, a measurement that starts at the arrival carries no silence: its noise floor cannot be measured.
Finally, no grade is a listening judgement. The assessment measures neither distortion nor pre-echo: those quantities are not in the measurements it has. And it never takes part in the equalisation — it reports, it does not correct.
A grade without its source is a decree. Every scale therefore carries its own: a quoted standard, or the date and the size of the body of measurements that calibrated it. The rule that governs them fits in two sentences. A bound taken from a standard does not move, even if most installations fall on the wrong side of it: what gets written down is that finding, not a new bound. A bound taken from a body of measurements is dated, and only moves through a written decision.
The three delay scales are read as fractions of one period at the frequency being judged — a quarter, a half, a whole period — with a floor of 9.5 ms: below that there is nothing left to correct, and a tighter bound would only describe itself.
One reading of the column needs a word of caution: the midrange decay is judged inside a window — a room that is too dull falls out of it from below, a room that is too reverberant from above.
Once the previews look correct, open the Finalization area to prepare the transfer of the calibration into the receiver. The transfer goes directly through the bridge: there is no file to generate and no external program to run.
At the top of the area, the Speaker Preset row shows the active speaker preset of the receiver (read live through the bridge) and lets you switch between Preset 1 and Preset 2: pick the one that will receive the calibration before transferring. On receivers without speaker presets, the row says so and the buttons are hidden.
Each preset has its own speaker setup. RCH reads the receiver again as soon as the preset changes. If the new preset does not have the same channels as the calibration, two subwoofers instead of four for instance, RCH says so and keeps your session as it is: switch back to the calibration's preset, or set the channels up identically on the receiver's current preset. The transfer refuses a calibration that does not match the receiver, without sending it anything.
The LPF for LFE field is decided here, and nowhere before: no other command writes it. In particular, Find Sub Alignment does not touch it — it aligns, and the LFE low-pass has its own button, right next to the field. It is the last step of the workflow (step 10), to be done once the crossovers are settled.
The Find Best LFE Low-Pass button, next to the LPF for LFE field, picks this value automatically. RCH tests every LFE low-pass frequency available for your AVR: for each one, it simulates the filter (LR24, the same type as the bass management low-pass) on the real sum of the subwoofers, takes into account the time shift this filter imposes on the LFE signal, then measures how well the result adds up with the complete front channels (FL, C, FR) — the part played by the speaker above its crossover AND its bass redirected to the subwoofers — across the whole useful LFE band, up to 120 Hz. The selected frequency is the one giving the most constructive summation; a low value that would cut content below 120 Hz is penalised exactly like a cancellation. The per-candidate details (mean summation loss, worst dip, time shift taken into account) are visible in the logs, along with the comparison against the previous value. The speaker filters and the sub sum must be ready (same prerequisites as Find Best Crossover); no other setting is modified.
Calibration filter generation is fully internal: RCH reads the filters present in REW (including any you regenerated or adjusted directly in its interface) and computes the impulse response in your AVR's format by itself. Supported filter types are PK, all-pass, low/high pass (LP, HP, LP Q, HP Q, LP1, HP1), notch (Notch, Notch Q), the low/high shelves (LS, HS, LS 6dB, LS 12dB, HS 6dB, HS 12dB, LS Q, HS Q), modal filters, and the Low pass and High pass crossovers in Butterworth (6 to 48 dB/oct) or Linkwitz-Riley (12, 24, 36 or 48 dB/oct) — each reproduces REW's output exactly. You can therefore place your own filters in REW's X1 and X2 slots: RCH reads them and folds them into the calibration, and never writes anything there itself.
Three shapes remain outside that list and stop the save with a message naming the slot concerned (X1, X2, or its number): Bessel (BE) crossovers, whose exact convention is not characterised; Linkwitz-Riley set to a slope that is neither 12, 24, 36 nor 48 dB/oct — REW accepts those and echoes them back unchanged, but actually realises the next lower slope in that list; and the BP, Bandpass, Low shelf and High shelf types. Replace or disable the filter concerned, then retry.
The transferred filters of every bass-managed speaker bake in an electrical Butterworth 12 dB/oct high-pass at the channel's crossover frequency. Cascaded with the BW12 the AVR applies, the junction with the subwoofer becomes a symmetric, in-phase LR24|LR24 pair — the ~2 dB bump at the handoff disappears (see step 9). Subwoofers and Large speakers are untouched, and each channel's level is preserved. The affected channels are listed in the logs. Note for MultEQ (non XT/XT32): with a crossover below 60 Hz the filter response is slightly truncated by the AVR's limited filter length (a warning is logged).
An Audyssey receiver stores two filter banks, Reference and Flat, which you switch between while listening in the Audyssey menu of the receiver. RCH lets you fill each of the two with your own calculations. Key point of the Audyssey format: the receiver only carries one set of delays, levels, distances and crossovers, shared by both banks — only the filters differ from one bank to the other.
The recommended path:
Save an .mdat before recomputing — in REW, File / Save all measurements. A regenerated preview replaces the one for the same channel, and the session file carries no signal data: the state of the Reference bank is nowhere else. The usual practice is to keep two of them, a reference .mdat and a Flat .mdat, and to reload in REW whichever one you want to look at again.
The Duplicate to the other bank button is a shortcut: it copies the already saved bank into the other one, to transfer the same filters on both sides without recomputing. Clear banks empties both banks to start over.
The transfer requires both banks to be saved. Before transferring:
Above the button, the panel shows a recap of what is actually going to be sent: the target receiver (model, MultEQ version, channel count), both banks with their target curve and save date, and, if the session was calibrated by someone else, their name. A warning appears when the banks no longer match the working state on screen. Another one appears when several subwoofers are declared and only one was measured: it names the measured subwoofer and the ones that will receive its calibration as is — same filter, same distance, same polarity, same trim level written into the receiver. The gain knob on the subwoofer itself is not touched by the transfer. That warning states a fact about the transfer, not a limitation of the receiver: it also appears on a receiver perfectly able to measure subwoofer by subwoofer when a single subwoofer measurement is left.
A single button, Transfer to AVR. It starts by having the receiver check the complete calibration without writing anything: if it does not match the actual configuration (active channels, amp assignment, MultEQ version), the transfer stops there, no setting is touched, and the error message says what does not match. Otherwise the real transfer follows straight on: progress is shown per channel, and Cancel transfer lets you interrupt — cancellation can take a few seconds if the receiver is finalizing. The amp assignment is read live from the receiver when the archive is built, so there is no possible mismatch between the transfer and the actual state of the receiver. If delays, levels or crossovers changed since a bank was saved, RCH flags that bank as stale: save it again before transferring.
Once the transfer completes, check the result by ear and in the receiver menu: MultEQ active, and switching between Reference and Flat available.
The Diagnostic drawer gathers the two files that play no part in the transfer. Download .rch.json downloads the calibration archive actually sent to the bridge: it is something to attach to a bug report. Create Settings File generates a full text record of the settings used. Neither can be re-imported, and neither is a calibration to share — to hand work over, use the session file (see Session).
The collapsible Session card of the Application page saves your work so you can resume it later. Export session downloads a .json file containing the settings, the working state of the measurements (without signal data) and the filter banks. Import session restores that file.
RCH also saves your state continuously in the browser (auto-save): a simple page reload loses nothing. The session file is for archiving a project, resuming it on another machine, or after a browser cleanup.
The measurements themselves live in REW: save them as a .mdat file (File / Save all measurements). To resume a project, reload the .mdat in REW first, then import the session: each measurement of the session is re-attached by identifier to the matching REW measurement. Session measurements that cannot be found in REW are reported (title and position) — reload the right .mdat and import again if that list is not empty.
The session file carries no machine-specific setting: not the REW address, not the bridge address, not your receiver's IP. You can send it to someone without imposing anything on them and without publishing your network.
The existing side exports (subwoofer measurements for MSO, CamillaDSP conversion) are kept and independent from the session file.
Someone from the community can run the calibration for you. You send your measurements, they do the work, they send back a file you transfer to your receiver. The session file is the only file travelling in both directions.
On the receiver owner's side (the one with the bridge):
On the calibrator's side (they do not have your receiver):
Back on the owner's side:
You do not need to reload the .mdat to transfer: the transfer is built from the filter banks, not from the measurements.
Do not share the .rch.json file from the Diagnostic drawer: it is a technical record meant for bug reports, it cannot be re-imported, and it does not replace the session file.
One card per fault: the symptom as the title, the likely cause under it, then the steps in order.
The log at the bottom of the page is the record of everything RCH decides. It is written for you, in the language of the interface, and it holds only four kinds of line:
So a line about an anomaly is always a full sentence, never a code. A refusal reads like this:
"1: FL: no filter could be computed over 20-200 Hz. This measurement is left uncorrected. Check its measurement and the target curve loaded in REW, then run it again; if the gap comes from a deep dip, equalisation will not fill it."
And a warning:
"SW1_P3: its impulse clips (1.14 above digital full scale). The measurement is clipped, so any filter computed on it will be wrong. Measure it again with a lower output level."
The detail of the calculation — phases, scores, parameters tried, the full report of the subwoofer optimizer — is not dropped for all that: it lives one level below. Set the log level to Debug to see it. That is also where the technical text of an error goes, with its whole chain of causes. The log export, for its part, carries everything whatever level is on screen: that is the file to attach to a bug report.
The five most talkative operations — multi-subwoofer optimizer, subwoofer sum equalization, speaker filter generation, Find Sub Alignment and the polarity check — are framed there: a rule carries the operation name where it starts, a dashed rule closes it and states how long it took. Everything the operation wrote sits between the two. Frames stay visible whatever log level you pick: even set to Error, you can see which operation produced the error.
Just before the closing rule, RCH publishes the operation's numeric summary. None of these values is recomputed for the occasion: they are the ones the calculation just produced. Depending on the operation, you read there:
Two controls to find your way.
myrch.fr. The browser reports that refusal exactly like an
offline bridge — hence the first step below.
http://localhost:7735/ in a new tab. If the bridge
dashboard appears, the bridge is running — and it is this site's
address that it refuses. If nothing appears, the bridge is not
running.
config.json file in
its data folder (%APPDATA%\rch-bridge\ on Windows,
~/Library/Application Support/rch-bridge/ on macOS,
~/.config/rch-bridge/ on Linux) containing
{"corsOrigins": ["https://myrch.fr", "http://localhost:7735",
"http://127.0.0.1:7735", "http://localhost:5173",
"http://127.0.0.1:5173", "http://localhost:4173",
"http://127.0.0.1:4173", "http://localhost:3000",
"http://127.0.0.1:3000"]}, then restart the bridge.
Copy the whole list: it replaces the binary's own
instead of adding to it, and leaving a line out cuts off the matching
address. The file must be in that folder, not next to the program. If
nothing changes, open rch-bridge.log in that same
folder: a malformed file is reported there, and nowhere else.
For more information, you can use the following resources:
If you notice a mismatch between this page and the application, trust the labels and behavior currently visible in the Application tab first.